Looks impressive but I can't escape the notion that surely some of the generated images will be very close to the some of the training images?

How am I to assess how original the generated results really are?